Nvidia introduced new open-weight AI models on Thursday designed for always-on agents in enterprise settings and edge computing. The Nemotron 3.5 Super VL is a 120-billion-parameter model optimized for enterprise agent tasks, delivering accuracy comparable to the larger Nemotron 3 Ultra. Additionally, Nvidia launched two lightweight Bolt models intended to run locally on AI PCs equipped with the Nvidia RTX Spark chip and DGX Spark workstations, now available through partners like Microsoft and Dell, according to techcrunch.com.
The Nemotron 3.5 Super VL model balances parameter count and performance, enabling enterprises to deploy capable agents without sacrificing speed or accuracy. The Bolt models target consumer and knowledge worker devices, allowing local AI processing on new PCs. Kari Briski, Nvidia’s vice president of Open Models and Software, highlighted that organizations can now choose models optimized for their infrastructure tier without compromising capability or deployability. Nvidia claims the Super VL outperforms competing models from Anthropic and OpenAI on agentic tasks after post-training, including document intelligence and security resilience, as detailed by techcrunch.com.
Nvidia’s launch addresses growing demand for AI models that can operate efficiently on both enterprise servers and edge devices. The open-weight approach contrasts with closed proprietary models, potentially accelerating adoption in sectors like finance, healthcare, and scientific research. Nvidia’s competitors include Anthropic and OpenAI, whose models Nvidia benchmarks against. The new models align with Nvidia’s broader strategy to integrate AI capabilities into hardware platforms like RTX Spark, reinforcing its position in the AI infrastructure market, according to techcrunch.com.
The new AI PCs powered by Nvidia’s RTX Spark chip and DGX Spark workstations featuring the Bolt models are currently entering the market through Microsoft and Dell partnerships, marking a key milestone in Nvidia’s AI hardware and software integration. Nvidia’s open models are available immediately, supporting enterprises and developers aiming to deploy AI agents across various environments, as reported by techcrunch.com.